You are on page 1of 6

SICE Annual Conference 2012

August 20-23, 2012, Akita University, Akita, Japan

Simplified Flow Rate Estimation by Decentralization of Kalman Filters in


Automatic Pouring Robot
Yoshiyuki Noda1 and Kazuhiko Terashima2
1
Department of Mechanical Engineering, University of Yamanashi, Yamanashi, Japan
(Tel: +81-55-220-8440; E-mail: noday@yamanashi.ac.jp)
2
Department of Mechanical Engineering, Toyohashi University of Technology, Aichi, Japan
(Tel: +81-532-44-6699; E-mail: terasima@me.tut.ac.jp)

Abstract: We describe here a method for estimating the pouring flow rate for tilting-ladle-type automatic pouring robot
used in casting industries. To precisely pour molten metal into the mold, controlling the flow rate of liquid flowing out of
the ladle is mandatory. However, it is difficult to directly measure the pouring flow rate by using a conventional flow meter,
because the flow meter is damaged by the high temperature molten metal. Thus, the pouring flow rate estimation system
using a state estimation theory has been proposed recently. In previous studies, the extended or the unscented Kalman
filters have been applied to the pouring flow rate estimation. The flow rate estimation by the extended Kalman filter can
be applied only the ladle with smooth shape. On the other hand, the flow rate estimation by the unscented Kalman filter
can be applied to the ladles with complicated shape. However, it is difficult to perform a real-time computation of the
unscented Kalman filter by the industrial controller. In this paper, the simplified flow rate estimation in which the Kalman
filters are decentralized to the motor systems and the pouring process is proposed. The effectiveness of the proposed flow
rate estimation method is demonstrated through simulations and experiments.

Keywords: Automatic Pouring Robot, Flow Rate Estimation, Kalman Filter

1. INTRODUCTION some ladles with complicated shape which is difficult to


derive the Jacobian matrix have been used in the actual
The pouring process in the casting industry requires plants. Thus, the flow rate estimation with the unscented
that molten metal be quickly and precisely poured into a Kalman filter has been proposed in [9]. However, it is
mold and not spill over the sprue cup[1][2]. In the present difficult to implement the flow rate estimation with the
automatic pouring robots, a teaching-and-playback ap- unscented Kalman filter to the industrial controller such
proach has mainly been applied to satisfy such require- as a PLC(Programable Logic Controller).
ments[3]. In this approach, the tilting-ladle motion gen- Therefore, in this paper, we propose a simplified pour-
erated by the operator is detected using a rotary en- ing flow rate estimation method for the casting pouring
coder fitted into the motor, and the process is recorded process. In the proposed estimation approach, the state
in the memory of a controller. Then, the automatic pour- estimation are decentralized to the motor systems and the
ing robot duplicates the memorized tilting-ladle motion. pouring process, respectively. In this study, we applied
However, this approach involves an immense amount of the proposed flow rate estimation method to the auto-
time and effort. Moreover, the present automatic pouring matic pouring robot with a cylindrical ladle. Then, the
robot cannot achieve a high degree of pouring precision effectiveness of the proposed method was demonstrated
due to changes in the pouring conditions. through the simulations and the experiments.
To precisely pour molten metal into the mold, con-
trolling the flow rate of liquid flowing out of the ladle 2. AUTOMATIC POURING ROBOT
is mandatory. In general, a liquid’s flow rate can be mea-
sured directly by a flow meter. However, since molten The tilting-ladle type automatic pouring robot is
metal’s temperature is much higher than 1300[deg], the shown in Fig.1. The automatic pouring robot can transfer
flow meter can be damaged by the molten metal. In the the ladle in two dimensions (Y- and Z-axes) and rotate
other approach, an outflow weight is measured by a load the ladle (Θ-axis). On the Θ-axis, the ladle is tilted by
cell installed in the pouring robot, and the pouring flow an AC servomotor. The drive torque of the AC servo-
rate can be obtained indirectly by differentiating the mea- motor can be amplified by reducing the gear ratio. The
sured outflow weight[4]. However, the measured out- rotation angle is measured by an encoder installed in the
flow weight includes much noise and is subject to the AC servomotor. The ladle is also transferred along the
load cell’s dynamics. Therefore, the pouring flow rate Y- and Z-axes of the automatic pouring robot by AC ser-
obtained from the outflow weight cannot be measured vomotors. The driving force of each motor is amplified
precisely. For solving the problem, flow rate estimation through a ball and screw mechanism. The ladle can be
using an extended Kalman filter has been proposed by independently transferred along each axis. The transfer
present authors[5]. This approach has been limited only distance on Y- and Z-axes are measured by an encoder
the ladle with smooth shape such as fan-shaped ladle, installed in each motor. Finally, a cylindrical ladle is used
because of requirement of Jacobian matrix. However, in this study. The cylindrical ladles have been applied to

-1465- PR0001/12/0000- 1465¥ 400 ©2012 SICE


Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.
Transfer Direction (Z-axis) Tilting Angle θ [deg] Ladle
Ladle Liquid Volume on Pouring
Motors and Liquid Height on Mouth Vr[m3]
Encoders Pouring Mouth h[m] Horizontal Area
Tilting Direction (Θ-axis)
on Pouring Mouth
Flow Rate
Pouring Mouth A[m2]
q [m3/s]
Sprue Cup
Mold Pouring Mouth Liquid Volume under
PouringMouthVs[m3]
Transfer Direction (Y-axis)
Fig. 3 Cross Section of Pouring Process
Width of Pouring Mouth
Fig. 1 Tilting-ladle-type Automatic Pouring Robot at Depth h b
Ladle
Lf (hb)[m]
Depth from
Input Voltage Angular Velocity Flow Rate Outflow Weight
ut ω q ρ wo Liquid Upper
PT PF Surface h b [m]
s
Motor for 1 Pouring Process
Tilting Ladle θ
s Angle
Initial Weight
ρ : Density of Liquid Outflow Liquid
-
of Liquid in Ladle Wi + Pouring Mouth
Input Voltage g-az
uz
Acceleration
az g Fig. 4 Geometric Parameters of Pouring Mouth
PZ
Motor for Outflow Weight
Measured by Load Cell
Transfer Ladle on Z-axis
wL
PL 3.2 Pouring Process Model PF
Load Cell Wi The model PF in Fig.2 represents the pouring process
Fig. 2 Signal Diagram of the Pouring Robot Model from the tilting angular velocity to the flow rate of the
outflow liquid and has been described in detail in [8].
A cross section of the pouring process is shown in
Fig.3, where θ is the tilting angle of the ladle, Vs (θ)[m3 ]
the practical automatic pouring robot.
is the volume under the pouring mouth, and A(θ)[m2 ] is
To measure the weight of the liquid in the ladle, a load
the horizontal area to the pouring mouth. The volume
cell is fitted to the arm supporting the ladle in the pouring
Vs (θ) and the area A(θ) depend on the tilting angle θ.
robot. The weight of the liquid in the ladle is obtained
Vr [m3 ] is the volume on the area A(θ), and h[m] is the
by subtracting the weight of the ladle, the motor, and the
height from the area A(θ) to the upper surface of the liq-
arm from the gross weight measured by the load cell.
uid in the ladle. When the liquid is above the lip of the
3. MODELING OF POURING ROBOT pouring mouth, the liquid is flowed out of the ladle with
the flow rate q[m3 /s].
The model of the pouring process in Fig.1 is repre- The mass balance of the liquid in the ladle in Fig.3 is
sented as the signal diagram shown in Fig.2. An input described as
voltage is applied to the motor for tilting the ladle, and dVr (t) ∂Vs (θ(t))
liquid in the ladle is poured into the mold by tilting the = −q(h(t)) − ω(t). (3)
dt ∂θ(t)
ladle. The weight of the liquid in the ladle is measured
by the load cell. Then, the outflow liquid can be obtained The liquid height h can be represented as
by subtracting the weight of the liquid in the ladle while Vr (t)
h(t) = . (4)
pouring from the initial gross weight on the load cell. On A(θ(t))
the other hand, the weight measured by the load cell is Using Bernoulli’s theorem, the flow rate q[m3 /s] at the
affected by the acceleration of the transferring the ladle liquid height h is
on Z-axis.  h(t) 
3.1 Motor Model PT q(h(t)) = c Lf (hb ) 2ghb dhb ,
0
The motor model PT in Fig.2 for tilting the ladle is (h(t) ≥ 0), (5)
described as a linear first-order system
where hb [m] is the depth from the upper surface of the
dω(t) 1 Kt
= − ω(t) + ut (t), (1) liquid as shown in Fig.4, and Lf (hb )[m] is the width of
dt Tt Tt the pouring mouth at the depth hb . In addition, c[-] is
where ω[deg/s] is the angular velocity of the tilting ladle, the flow rate coefficient, and g[m/s2 ] is the acceleration
ut [V] is the input voltage applied to the motor, Tt [s] is of gravity. By substituting Eq.(4) to Eq.(3), the nonlinear
the time constant, and Kt [deg/sV] is the gain. The tilting pouring process model PF can be derived as
angle θ[deg] is given as dh(t) q(h(t)) 1
=− −
dθ(t) dt A(θ(t)) A(θ(t))
= ω(t) (2)  
dt ∂A(θ(t)) ∂Vs (θ(t))
× h(t) + ω(t). (6)
and can be measured by an encoder. ∂θ(t) ∂θ(t)

-1466-
Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.
Pouring Robot Pouring Robot
ut ω wo ut ω wo wL
PT θ PF PL wL PT θ PF PL
Motor for Motor for Pouring
Pouring Tilting Ladle Process
Tilting Ladle Process
az uz az
uz PZ PZ xz
xz Motor for
Motor for Transferring
Transferring Ladle on Z-axis
Ladle on Z-axis
Linear State az
q
Estimation Nonlinear
State
q
State Estimation Linear State ω Estimation
Estimation
Fig. 5 Block Diagram of Flow Rate Estimation in Previ- Fig. 6 Block Diagram of Proposed Flow Rate Estimation
ous Studies

from the pouring robot are the tilting angle θ of the ladle,
The actual weight wo [kg] of the liquid flowing out of the position xz of the ladle on Z-axis, and the outflow
the ladle is represented as weight wL measured by the load cell. The flow rate q
dwo (t) can be estimated from input and output signals by using
= ρq(h(t)), (7) the state estimation method. The extended Kalman fil-
dt
ter[5] or the unscented Kalman filter[9] is applied as the
where ρ[kg/m3 ] is the density of the liquid in the ladle. state estimation method. Since the pouring process has
3.3 Load Cell Model PL the nonlinear characteristics as seen from Eqs.(5) and (6),
The load cell dynamics PL in Fig.2 is modeled as a the extended or the unscented Kalman filter as the nonlin-
linear first-order system ear state estimation is used to the flow rate estimation in
the pouring robot.
dwL (t) 1 However, the motor systems PT and PZ can be repre-
= − wL (t)
dt TL sented as the linear system as shown in Eqs.(1) and (9).
1 (Wi − wo (t))az (t) Furthermore, the motor systems are not influenced from
+ {wo (t) − }, (8)
TL g the pouring process and the load cell. Therefore, the lin-
where wL [kg] is the weight of the outflow liquid mea- ear state estimation is applied to each motor system for
sured by the load cell, and TL [s] is the time constant rep- estimating the angular velocity of the tilting ladle and the
resenting the response of the load cell. az [m/s2 ] is the ac- acceleration of the transferring the ladle on Z-axis. Then,
celeration of transfer of the ladle on Z-axis, and Wi [kg] using the nonlinear state estimation, the flow rate can be
is the initial gross weight on the load cell before pouring. estimated from the estimated angular velocity and the es-
timated acceleration on Z-axis. The block diagram of the
3.4 Motor Model PZ proposed flow rate estimation is shown in Fig.6. In this
The motor model PZ in Fig.2 for transferring the ladle paper, a steady state Kalman filter as the linear state esti-
on Z-axis is described as a linear first-order system mation is applied to each motor system, and an extended
Kalman filter as the nonlinear state estimation is applied
dvz (t) 1 Kz
az (t) = = − vz (t) + uz (t), (9) to the pouring process with the load cell.
dt Tz Tz The steady state Kalman filter for estimating the an-
where vz [m/s] is the velocity of the transferring ladle, gular velocity of the tilting ladle is constructed with the
uz [V] is the input voltage applied to the motor, Tz [s] is state and output equations as
the time constant, and Kz [m/sV] is the gain. az [m/s2 ]    1    Kt 
d ω 0 ω
is the acceleration of the transferring the ladle on Z-axis. = Tt + Tt ut , (11)
The position xz [m] is given as dt θ 1 0 θ 0
 
 ω
dxz (t) y= 0 1 . (12)
= vz (t), (10) θ
dt
and can be measured by the encoder fitted to the motor The estimated angular velocity ω̄[deg/s] by this state es-
for transferring the ladle on Z-axis. timation is applied to the state estimation of the pouring
process.
4. FLOW RATE ESTIMATION And, the steady state Kalman filter for estimating the
acceleration of the transferring ladle on Z-axis is con-
In previous studies[5][9], the flow rate estimation has structed with the state and output equations as
been constructed with the structure as shown in Fig. 5. In    1    Kz 
d vz T 0 vz
Fig. 5, the block diagram related to the pouring robot is = z + Tz uz , (13)
illustrated briefly. The input signals to the pouring robot dt xz 1 0 xz 0
 
are the input voltages ut and uz to the motors for tilting  vz
y= 0 1 . (14)
and transferring ladle, respectively. The output signals xz

-1467-
Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.
The acceleration āz [m/s2 ] can be estimated by substitut- Lp=0.175[m] Lf =0.05[m]
Ladle
ing the estimated velocity v̄z [m/s] and the input uz to
1 Kz
āz (t) = − v̄z (t) + uz (t). (15) Pouring
Tz Tz Mouth
The extended Kalman filter for estimating the flow rate
is constructed with the state and output equations as Hf=
0.3[m] Hb=
dx 0.2[m]
= f (x), (16)
dt D=0.24[m]
y = wL , (17)
where, Lb =0.09[m]
 T
x= h wo wL , Fig. 7 Geometory of Cylindrical Ladle
⎡ ⎤
− q(h) − A(θ)
1 −3
x 10
⎢ A(θ) ∂A(θ)  ⎥
0.075
(a)Area A
10
(b)Volume Vs
⎢ × h + ∂Vs (θ)
ω̄ ⎥
⎢ ⎥

Volume [m ]
∂θ ∂θ 0.07 8

3
Area [m2]
f (x) = ⎢
⎢ ρq(h)
⎥.

⎢ 1 1 ⎥ 0.065 6
⎣ − TL wL (t) + TL {wo (t) ⎦
(Wi −wo (t))āz 0.06 4
− g }
0.055 2
30 40 50 60 30 40 50 60
And, the Jacobian matrix used for updating the Kalman Angle[deg] Angle[deg]
gains in the extended Kalman filter is shown as Fig. 8 Model Parameter’s of Ladle
⎡ ⎤
α 0 0
∂f (x) ⎣ 11
= α21 0 0 ⎦, (18)
∂x 0 α α 32 33 order state vector on a real-time by the industrial con-
troller. In the proposed estimation approach, the Kalman
where,
gains related to the motor systems can be calculated pre-
1 ∂q(h) ∂A(θ) ∂q(h) liminary using the steady state Kalman filter. And, the
α11 = − ( + ω̄), α21 = ρ ,
A(θ) ∂h ∂θ ∂h Kalman gains to 3rd order state vector in the pouring pro-
1 āz 1 cess can be calculated on a real-time using the extended
α32 = (1 − ), α33 = − . Kalman filter. The calculated load of the flow rate esti-
TL g TL
mation is reduced using the proposed approach.
The derivative of the flow rate q to the liquid height h can
be derived from Eq.(5) as 5. SIMULATIONS
∂q 
= cLf (hb ) 2ghb . (19) The proposed flow rate estimation which is shown in
∂h the previous section is performed in the simulation of au-
Then, the flow rate q̄[m3 /s] can be estimated by substi- tomatic pouring robot. The cylindrical ladle used in this
tuting the estimated liquid height on the pouring mouth study is shown in Fig.7. From the ladle’s shape as shown
to in Fig.7, the volume under the pouring mouth Vs (θ)[m3 ]
 h̄  and the horizontal liquid area to the pouring mouth A[m2 ]
q̄(h̄) = c Lf (hb ) 2ghb dhb . (20) are obtained as shown in Fig.8. In Fig.8, (a) represents
0
the horizontal liquid area to the pouring mouth, and (b)
The Jacobian matrix shown in Eq.(18) does not re- shows the liquid volume under the pouring mouth. And,
quire the second order derivative of the model parame- the other model parameters are shown in Table 1.
ters A(θ) and Vs (θ) to the tilting angle θ. In the previ- For the state estimation of the motor system for tilting
ous study[5], the model parameters have to be satisfied as ladle, the covariances of system noise and measurement
A(θ), Vs (θ) ∈ C 2 for constructing the flow rate control noise are assumed as
using the extended Kalman filter. Therefore, this flow rate
estimation cannot be applied to the practical automatic Qt = diag(1 × 10−6 [deg2 /s2 ], 1[deg2 ]) × 10−9 ,
pouring robot using the ladle with complicated shape. Rt = 1 × 10−15 [deg2 ],
The proposed flow rate estimation can be applied to the where Qt and Rt mean the covariances of the system
practical automatic pouring robot, since the Jacobian ma- noise and measurement noise, respectively. And, for the
trix in this approach does not require A(θ), Vs (θ) ∈ C 2 . state estimation of the motor system for transferring ladle
Moreover, in the previous studies[5][9], 7th order state on Z-axis, the covariances of system noise and measure-
vector was estimated by the nonlinear state estimation. In ment noise are assumed as
the nonlinear state estimation such as extended and un-
scented Kalman filter, the Kalman gains are updated on a Qz = diag(1 × 10−6 [m2 /s2 ], 1[m2 ]) × 10−9 ,
real-time. Therefore, it is difficult to estimate the higher Rz = 1 × 10−15 [m2 ].

-1468-
Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.
Table 1 Model Parameters of Automatic Pouring Robot Ideal Tilting Angle
at Beginning of Outflow
θbc [deg] Ladle
Time Constant of Motor for
Tilting Ladle, Tt [s] 0.60 ×10−2
Ideal Liquid Volume
Gain of Motor for Tilting Ladle Actual Liquid Volume
Kt [deg/sV] 24.58
Time Constant of Motor for
Transferring Ladle on Z-axis, Tz [s] 0.60 ×10−2
Gain of Motor for Transferring Ladle Pouring Mouth
on Z-axis Kz [m/sV] 24.58
Flow Rate Coefficient of Pouring, c[-] 0.79 Fig. 9 Tilting Angle of Ladle at the Beginning of Outflow
Density of Liquid (Water), ρ[kg/m3 ] 1 × 103
0.50 ×10−1

Angular Velocity [deg/s]


Time Constant of Load Cell, TL [s] Propose
0.5 10
Variance of Measurement Noise (a) UKF
in Encoder to Tilting Angle, [deg2] 0.1 ×10−14

Input [V]
Sim
0 0
Variance of Measurement Noise (b)
in Encoder to Position on Z-axis, [m2 ] 0.1 ×10 −14

Variance of Measurement Noise −0.5 −10


0 5 10 0 5 10
in Load Cell, [kg2 ] 0.18 ×10 −2
55 0.04

Liquid Height [m]


(c) (d)

Angle [deg]
50
45 0.02
Finally, the covariances of system noise and measurement 40

noise to the state estimation of the pouring process are 35

Outflow Weight by Loadcell [kg]


0
assumed as 0 5 10 0 5 10
Outflow Weight [kg]

3 3
Qf = diag(3 × 101 [m2 ], 1[kg2 ], 1[kg2 ]) × 10−9 ,
Rf = 1.82 × 10−3 [kg2 ].
2 (e) 2 (f)
1 1
In the covariances of the system noise, since the liquid
height h is the most uncertain in all of the state variable, 0 0
0 5 10 0 5 10
the system noise with respect to the liquid height has been Time [s]

increased. Fig. 10 Simulation Results of State Estimation


5.1 Simulation Results
−3
The proposed flow rate estimation is implemented to 1
x 10

the automatic pouring robot shown in Fig.1. In this sec- Propose


0.8 UKF
Flow Rate [m3/s]

tion, the effectiveness of the proposed flow rate estima- Sim


0.6
tion is verified in simulations. For comparison, the flow
rate estimation using the unscented Kalman filter is also 0.4
performed in the simulations. The design of the un- 0.2
scented Kalman filter is referred to [9]. The model pa- 0
rameters in the unscented Kalman filter are the same as 0 2 4
Time [s]
6 8 10

the proposed approach.


Fig. 11 Simulation Results of Flow Rate Estimation
In the simulation, the actual tilting angle at the begin-
ning of outflow is larger than the ideal one as shown in
Fig.9. In practical use, it is caused frequently that the
tilting angle at the beginning of outflow calculated in the
are depicted. There, the blue solid lines are the results of
controller is different from that in actual pouring process.
the proposed state estimation. The green solid lines result
In the simulation, the ideal tilting angle θbc [deg] at the
by the state estimation using the unscented Kalman filter.
beginning of outflow is set as θbc = 35[deg], however
The red dashed lines are the state variables on the pouring
the actual tilting angle θba [deg] is as θba = 36[deg]. The
process model as shown in Eqs.(1)-(8).
ideal tilting angle has been set preliminary into the flow
rate estimations. And, the results of the flow rate estimation are shown
The simulation results of the state estimation are in Fig.11. The line colors are also same as Fig.10. As
shown in Fig.10. In the diagrams (a), (b), (c), (d), (e) seen from the simulation results, accuracy of the flow rate
and (f), the control input ut , the tilting angular velocity estimation by the proposed approach is similar to that by
ω, the tilting angle θ of the ladle, the liquid height h on the unscented Kalman filter. Therefore, the proposed flow
the pouring mouth, the weight of outflow liquid wo and rate estimation can be constructed simply and estimate
the weight wL of outflow liquid measured by the load cell the flow rate precisely.

-1469-
Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.
7. CONCLUSIONS

Angular Velocity [deg/s]


0.5 10
(a) Propose We proposed the flow rate estimation system to the
Input [V]

Sim
0 0
automatic pouring robot in casting industry. In order to
(b) implement the flow rate estimation system to the indus-
trial controller, the state estimation is decentralized to the
−0.5 −10
0 5 10 0 5 10 motor systems and the pouring process. Since the motor
50 0.03 system is represented as the linear model, the linear state
Liquid Height [m]
(c) Propose
estimation is constructed to the motor system. Then, the
Angle [deg]

0.02 Sim
40 (d)
flow rate can be estimated by applying the estimated an-
0.01 gular velocity and velocity of the transfer on Z-axis into
the nonlinear state estimation with the pouring model.
Outflow Weight by Loadcell [kg]

30 0
0 5 10 0 5 10
The simulation and experimental results showed that the
Outflow Weight [kg]

3 3 flow rate of the outflow liquid can be precisely estimated


(e) Propose
2 2 Exp using the proposed estimation system.
The proposed flow rate estimation can be applied to
1 1 (f)
the automatic pouring robots with various ladles, and
0
0 5 10
0
0 5 10
constructed simply. Therefore, the proposed estimation
Time [s] method is useful in practical use.

Fig. 12 Experimental Results of State Estimation REFERENCES


x 10
−4
[1] W. Lindsay, “Automatic Pouring and Metal Distribu-
8
Propose tion Systems”, Foundry Trade Journal, February, no.
Flow Rate [m3/s]

6 Sim 6, pp 151–176, 1983.


[2] K. Terashima, “Recent Automatic Pouring and
4
Transfer Systems in Foundries”, Materials Process
2 Technology (in Japanese), vol. 39, no. 6, pp 1–8,
1998.
0
0 2 4 6 8 10 [3] J. Watanabe and K. Yoshida, “Automatic Pouring
Time [s]
Equipment for Casting - Mel Pore System”, Indus-
Fig. 13 Experimental Results of Flow Rate Estimation trial Heating (in Japanese), vol. 29, no. 4, pp 19–27,
1992.
[4] K. Shinohara and H. Morimoto, “Development of au-
6. EXPERIMENTS tomatic pouring equipment”, Journal of the Society
of Automotive Engineers of Japan in Japanese, vol.
The proposed flow rate estimation has been applied to 46, no. 11, pp 79–85, 1992.
the automatic pouring robot in our laboratory as shown [5] Y. Noda, Y. Matsuo, K. Terashima and Y. Zheng, “A
in Fig.1. For safety reason, the target liquid is the water. Novel Flow Rate Estimation Method Using Extended
The model parameters are same as the simulations in pre- Kalman Filter and Sensor Dynamics Compensation
vious section. In the experimental conditions, the tilting with Automatic Casting Pouring Process”, Preprints
angle at the beginning of outflow is same as the simula- of 17th IFAC World Congress, pp 710–715, 2008.
tion conditions. However, the beginning angle of tilting [6] R. Kandepu, B. Foss and L. Imsland, “Applying the
ladle is 30[deg] in this experiment. Unscented Kalman Filter for Nonlinear State Estima-
The experimental results of the state estimation are tion”, Journal of Process Control, vol. 18, pp 753–
shown in Fig.12. The layout in Fig.12 is same as Fig.10. 768, 2008.
In diagrams (b), (c), (d), (e) and (f), the blue solid lines [7] G. Terejanu, T. Singh and P. D. Scott, “Unscented
are the results of the proposed state estimation. The green Kalman Filter/Smoother for a CBRN Puff-Based
dashed lines are the state variables in simulation of the Dispersion Model”, Proceedings of the 10th Interna-
pouring process model. In diagram (f), the green solid tional Conference on Information Fusion, 2007.
line is the experimental result of the weight of the out- [8] Y. Noda and K. Terashima, “Modeling and Feed-
flow liquid measured by the load cell. And, the results of forward Flow Rate Control of Automatic Pouring
the flow rate estimation are shown in Fig.13. In the ex- System with Real Ladle”, Journal of Robotics and
periment, the tilting angle at the beginning of outflow in Mechatronics, vol. 19, no. 2, pp 205–211, 2007.
the flow rate estimation is set 35[deg], however the liquid [9] Y. Noda and K. Terashima, “Flow Rate Estimation
begins to flow out of the ladle at the tilting angle 36[deg]. Using Unscented Kalman Filter in Automatic Pour-
Even if there is the error of the tilting angle at the be- ing Robot”, Proceedings of SICE Annual Conference
ginning of outflow in the flow rate estimation, the flow 2011, pp 769–774, 2011.
rate can be estimated precisely by using the proposed ap-
proach.

-1470-
Authorized licensed use limited to: ULAKBIM UASL - Sakarya Universitesi. Downloaded on February 25,2024 at 17:15:37 UTC from IEEE Xplore. Restrictions apply.

You might also like